A similarity-based range-frequency model for two-category rating data

Abstract

638A well-known feature of category ratings is their dependence on the contextual distribution of stimuli being judged. For example, a target person will receive higher ratings on such dimensions as attractiveness, likeableness, or sociability if the ratings of a recently encountered set of individuals are primarily low on those dimensions (for a review, see Wedell, Hicklin, & Smarandescu, 2007). Although these context effects are quite general and extend to judgments along abstract or complex dimensions, they have been tested most rigorously in the domain of psychophysical judgment, with its careful control of stimulus attributes and context. Such tests have demonstrated strong support for Parducci's (1965) range-frequency theory of judgment over historically competing models based on one or two contextual referents (Niedrich, Sharma, & Wedell, 2001).Range-frequency theory proposes that two principles are responsible for contextual shifts in evaluations. According to the range principle, judgments describe the dimensional location of the target relative to the subjective minimum and maximum values of the activated contextual distribution. Thus, presenting extreme stimuli that extend the range upward or downward can have profound effects on judgment via the range principle. However, strong context effects can be achieved when the stimulus range is equated across contexts but the relative ranks are manipulated. The frequency principle accounts for these effects by positing that judgments reflect the relative rank within the activated contextual distribution. Following the frequency principle, the mean rating function tends to mimic the percentile rank function for a given distribution.Historically, tests of range-frequency theory have focused on predicting mean ratings and largely ignored predictions of category usage. One exception is Parducci's (1965) range-frequency limen model, in which the locations of category boundaries (limens), determined by range-frequency principles, were placed within a Thurstonean categorization framework that predicted proportion of category usage. This approach assumed that range-frequency principles guided placement of category boundaries, whereas stimulus discrimination for pairs common to each context remained context independent. Although the model provided reasonable qualitative fits in early tests, subsequent quantitative tests of the model suggested its inadequacy (Parducci & Perrett, 1971), which appeared to be due largely to the model's false assumption that discriminability is independent of context.Although one may account for context effects on judgment and discrimination through distinct theoretical mechanisms, it is intriguing to consider how a common mechanism may serve to explain both effects. The rangefrequency theoretical framework offers such a mechanism if one assumes that judgment is driven by similarity comparisons on an implicit range-frequency scale. In support of this assertion, Krumhansl (1978) has demonstrated that stimulus densities can increase dissimil...

Categorization reduces the effect of context on hedonic preference

“…Our experiment evaluated preference judgments, which are essentially judgments of hedonic difference. Previous studies of range effects have focused on discriminability rather than on difference, and those studies found that decreasing the stimulus range increases discrimination (size of squares- Parducci & Perrett, 1971;Wedell, 2008;loudness-Gravetter & Lockhead, 1973;Lockhead & Hinson, 1986; but see Parker, Murphy, & Schneider, 2002). We know of only one study showing an effect of categorization on condensation that resembles the effect that was demonstrated in ours: a study of loudness discrimination that procedurally separated loud and soft tones at a single frequency.…”

Increased hedonic differences despite increases in hedonic range

“…More important for the present purposes, there are also many theories (e.g., Durlach & Braida, 1969;Gravetter & Lockhead, 1973;Lockhead & Hinson, 1986;Luce, Green, & Weber, 1976;Parducci, 1995;Petzold, 1990;Sarris, 2006;Wedell, 2008) predicting that expanding the perceptual range of stimuli will result in reduced discriminability or difference between any pair of stimuli within the range (see, e.g., Ward, Armstrong, & Golestani, 1996, for some examples and a thorough discussion of the issues). So one might expect that condensation should accompany contrast-that expansion of the hedonic range with hedonically positive or negative context stimuli should result in a reduction of preference between the test stimuli.…”

Context effects on reproduced magnitudes from short-term and long-term memory

“…Parducci (1995) has pointed out that the frequency principle in RFT works to maximize information transmitted about stimuli within a set, with enhancement of differences in denser regions of the distribution. These enhanced differences have been shown to result in decreased pairwise similarity (Wedell, 1996) as well as increased discriminability (Wedell, 2008). It is therefore reasonable to assume that when learning identities of different stimuli, rank information will be emphasized and encoded in memory.…”